Navigation Performance and Integrity Analysis in Decentralized PNT Systems

Abstract

Modern Position, Navigation and Timing (PNT) systems are driven by the need for automation and precise positioning in challenging modern applications such as self-driving cars, UAV package delivery etc. As systems grow in complexity, traditional centralized filtering techniques face limitations in efficiently handling vast sensor data. Further, modern PNT systems involve not only sensor fusion but also system fusion. This drives the need to have a scalable, flexible and modular navigation architecture that can fuse vast sensor data and multiple systems while accommodating the rising PNT performance demands. The dissertation investigates decentralized estimation as a solution to this challenge. It presents techniques for integrating decentralized filtering systems and addresses the challenges of ensuring performance in black-box systems. The study specifically scrutinizes issues related to accuracy, integrity, continuity, and fault detection performance. It reveals that insufficient knowledge about the level of information correlation between fused systems can adversely affect integrity and fault detection, consequently diminishing system safety and reliability. The thesis introduces methods to tackle both fault-free and faulted integrity performance issues that arise when fusing numerous disparate sensors and systems with unknown or poorly-modeled inter-system correlations. It details Monte Carlo simulation results and experimental validation through UAV flight test data to demonstrate the efficacy of the proposed approach for dealing with unknown correlations. The insights into performance trade-offs and the techniques proposed in this work contribute to the understanding of the requirements for designing decentralized systems, especially for safety-critical applications subject to stringent certification and safety standards.